Use Of XGBoost Method For Very Short-Term Radiation Forecasting on Adaptive Solar Cells

Abstract

Solar radiation constitutes a pivotal determinant influencing the efficacy of solar panels, wherein fluctuations may engender uncertainty regarding the generated power and subsequently affect the stability and dependability of solar power systems. The objective of this research is to formulate a method for the short-term prediction of solar radiation employing the XGBoost algorithm. This methodology encompasses data pre-processing, the implementation of the XGBoost model, and the evaluation of the model utilizing the RMSE, MSE, MAPE, and MAE metrics. The experimental findings reveal that the predictive model exhibits commendable performance, with MAE values recorded at 3.3823, MSE at 4001.60, RMSE at 63.26, and MAPE at 5.09%. Despite the presence of discrepancies between the predicted outcomes and the actual data, the overarching trend in the data demonstrates a commendable level of accuracy, with a mere deviation of 5%. These results suggest that the integration of XGBoost methodologies has the potential to enhance the precision of solar radiation predictions for adaptive solar panel systems. Subsequent research endeavors are anticipated to cultivate more robust models through the utilization of an expanded dataset and alternative machine learning combinatorial models to further refine predictions amidst diverse meteorological conditions.